Yılmaz et al. (2026) Neural circuit policy and hybrid deep learning models for enhanced meteorological drought forecasting performance

Identification

• Journal: Applied Soft Computing
• Year: 2026
• Date: 2026-02-17
• Authors: Mustafa Utku Yılmaz, Talha Burak Alakus
• DOI: 10.1016/j.asoc.2026.114844

Research Groups

• Department of Civil Engineering, Kirklareli University, Kirklareli 39100, Türkiye
• Department of Software Engineering, Kirklareli University, Kirklareli 39100, Türkiye

Short Summary

This study introduces the novel Neural Circuit Policy (NCP) deep learning model for meteorological drought forecasting using the Standardized Precipitation Index (SPI) at multiple time scales, demonstrating its superior performance, especially when integrated into hybrid models, for both forecasting accuracy and drought category classification.

Objective

• To introduce and evaluate the Neural Circuit Policy (NCP) deep learning model for forecasting the Standardized Precipitation Index (SPI) at 3-, 6-, 9-, and 12-month time scales.
• To compare NCP's performance against established machine learning (Random Forest, eXtreme Gradient Boosting) and deep learning (Long Short-Term Memory, Convolutional Neural Network-Long Short-Term Memory) models.
• To investigate the potential of NCP-based hybrid models (NCP-Random Forest, NCP-eXtreme Gradient Boosting, Convolutional Neural Network-NCP, Long Short-Term Memory-NCP) for enhanced drought forecasting accuracy and drought category classification.

Study Configuration

• Spatial Scale: A drought-prone basin in Türkiye.
• Temporal Scale: Forecasting at 3-, 6-, 9-, and 12-month time scales.

Methodology and Data

• Models used: Neural Circuit Policy (NCP), Random Forest (RF), eXtreme Gradient Boosting (XGB), Long Short-Term Memory (LSTM), Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), NCP-RF, NCP-XGB, CNN-NCP, LSTM-NCP.
• Data sources: Meteorological data (e.g., precipitation, temperature) used to calculate the Standardized Precipitation Index (SPI).

Main Results

• NCP achieved high forecasting performance across all SPI time scales, with KGE > 0.92, NSE > 0.93, RMSE < 0.23, MAE < 0.15, and R2 > 0.94.
• The Borda Count confirmed that NCP consistently ranked highest among individual models, followed by LSTM, XGB, and RF.
• NCP-based hybrid models, particularly LSTM-NCP, demonstrated superior drought category classification accuracy (weighted F1 > 0.89).
• CNN-LSTM performed well but was generally outperformed by NCP and its hybrid configurations.

Contributions

• First application of the Neural Circuit Policy (NCP) deep learning model for meteorological drought forecasting.
• Demonstration of NCP's superior performance for SPI forecasting compared to established machine learning and deep learning models.
• Introduction and evaluation of novel NCP-based hybrid models, showing enhanced accuracy in both drought forecasting and drought category classification.

Funding

[No funding information was provided in the paper text.]

Citation

@article{Yılmaz2026Neural,
  author = {Yılmaz, Mustafa Utku and Alakus, Talha Burak},
  title = {Neural circuit policy and hybrid deep learning models for enhanced meteorological drought forecasting performance},
  journal = {Applied Soft Computing},
  year = {2026},
  doi = {10.1016/j.asoc.2026.114844},
  url = {https://doi.org/10.1016/j.asoc.2026.114844}
}